Network resource planning for evolvability in software-defined infrastructure

摘要

Software-defined infrastructure (SDI) makes it possible to construct virtual networks with finer time granularity than was possible before, possibly allowing immediate responses to customer requests to increase or decrease resource levels to match demand. Adaptation by virtual network embedding (VNE) control is a promising technique for handling demand fluctuations, and choosing a good physical resource design is important for adaptability of VNE control. Despite its importance, strategies to design a physical network that promotes VNE adaptability have not been previously explored. Here, we propose an SDI resource-design strategy that increases the diversity of VNE solutions by using a control system that varies with demand fluctuation. The proposed strategy imitates biological evolutionary adaptation in order to improve evolvability. We use the proposed strategy to construct a method for reinforcing the computational capacity of physical nodes, and conduct experiments by computer simulation. Our results shows that the probability of convergence with VNE control is improved by using the proposed physical-resource reinforcement, achieving a gain of up to 19% relative to a basis reinforcement.